Diagnostic tools for differentiating pleural mesothelioma from lung adenocarcinoma in paraffin embedded tissue. II. Design of an expert system and its application to the diagnosis of mesothelioma.

A panel of 14 antibodies (panepithelial antibody Lu-5, anti-keratin-18, anti-keratin-7, Ber-EP4, anti-Leu-M1, HEA-125, anti-carcinoembryonic antigen, anti-blood group-related antigens A, B, H, B72.3, anti-placental alkaline phosphatase, anti-vimentin and BMA-120), which have been evaluated for use in differentiating mesothelioma from lung adenocarcinoma, was applied to a group of 24 suspected mesotheliomas. Using the established qualitative, descriptive criteria derived from monovariate statistical analysis of the tumour control groups (definite mesotheliomas, adenocarcinomas), a definitive allocation was possible in only 25% of suspected cases. We therefore constructed two "expert systems", based on multivariate discriminant analysis with either the ALLOC 80 program for ordinal data or a newly developed analysis program for binomial data. With these two systems diagnostic allocation of suspected mesotheliomas was improved to 75% and 79%. The use of binomial data ("positive" versus "negative") in conjunction with the probability-based test system is of particular interest because the primary data are easy to record and the test results have a higher statistical probability.